3D Frequency-Difference Electrical Impedance Tomography for Early Subcutaneous Edema Detection Using Tikhonov Regularization and Physics-Informed Neural Networks

Abstract

Subcutaneous edema is difficult to detect at an early stage because the associated conductivity changes are subtle and hard to localize using conventional imaging methods. This study presents the 3D Frequency Difference-Electrical Impedance Tomography (FD-EIT) framework for the detection of early-stage edema in the lower leg region. A 3D lower-leg model with 16 electrodes was simulated, and the boundary voltages and differential measurements were generated under a broader set of synthetic edema scenarios representing varied anomaly extent, irregularity, conductivity, contrast, and measurement difficulty. Reconstruction was evaluated using two approaches: the baseline method using Tikhonov regularization and the Physics-Informed Neural Network (PINN) method. The results show that the baseline method produces stable but spatially diffuse reconstructions, whereas the PINN method provides more localized and structured anomaly patterns that are aligned more closely with the ground-truth edema regions. Quantitative evaluation with RMSE, MAE, relative error, and Dice score showed better reconstruction accuracy and edema-mask localization for the PINN compared to the baseline method. However, the results should be interpreted as a controlled synthetic proof-of-concept because the PINN was trained and evaluated using scenario-specific simulated data. Therefore, the reported localization performance does not represent clinical diagnostic accuracy. Future work will focus on realistic anatomical modeling, electrode contact effects, phantom validation, and experimental lower-leg measurements.